1. Field of the Invention
The present invention relates to a semiconductor memory device and a semi-finished product for an integrated semiconductor device. More particularly, the present invention relates to a semiconductor memory device that allows an acceleration test such as a burn-in test, and a semi-finished product for an integrated semiconductor device that allows such testing in a wafer state prior to dicing.
2. Description of the Background Art
FIG. 39 is a block diagram schematically showing a structure of a memory cell and a row decoder widely used in a dynamic random access memory device (referred to as DRAM hereinafter). Referring to FIG. 39, a memory cell 111 includes an access transistor 112 and a cell capacitor 113, connected to a corresponding word line WL and a bit line BL. A row decoder selecting one word line in response to a row address signal includes a row predecoder 121a, a predecode signal line 122, and a plurality of word drivers. One word driver is provided corresponding to each word line WL. In FIG. 39, only one word driver WD is typically shown. Row predecoder 121a predecodes row address signals RA1-RA4 and complementary row address signals /RA1-/RA4 to generate predecode signals X1-X8 which are supplied to predecode signal line 122. Each word driver is activated in response to one of predecode signals X1-X4 and one of predecode signals X5-X8. When the word driver is rendered active, a boosted potential VPP higher than the power potential VPP higher than the power supply potential is supplied to a corresponding word line.
Word driver WD includes N channel MOS transistors 124 and 125 connected in series between a precharge node NX and a ground node, P channel MOS transistors 126 and 127 connected in parallel between a boosting node to which boosted potential VPP is supplied and precharge node NX, and a P channel MOS transistor 128 and an N channel MOS transistor 129 forming a CMOS inverter. During an inactive period of the DRAM (chip), a precharge signal PR of an L level (logical low) is applied to the gate electrode of transistor 126. In response, precharge node NX is precharged to an H level (logical high). Therefore, all the word lines are fixed to an L level when the chip is inactive. When the chip is activated, precharge signal PR is pulled up to an H level from an L level, whereby the precharge operation of node NX by transistor 126 is suppressed. However, word line WL maintains its L level since node NX is held at the H level. This is because the potential of word line WL is fed back to the gate electrode of transistor 127, whereby transistor 127 continuously supplies charge to node NX. Therefore, the charge of node NX must be discharged towards the ground node in order to render word line WL active. This word driver WD has both transistors 124 and 125 turned on when one predecode signal DECA out of predecode signals X1-X4 attains an H level and one predecode signal DECB out of predecode signals X5-X8 attains an H level. As a result, the potential of node NX is pulled down towards the L level, whereby transistors 128 and 129 are turned on and off, respectively. Thus, word line WL is activated, so that the potential thereof is boosted to the level of the boosted potential VPP.
FIG. 40 is a circuit diagram showing a structure of row predecoder 121a shown in FIG. 39. Referring to FIG. 40, row predecoder 121a includes NOR gates 1211-1218, inverters 1221a-1228a and 1231-1238. Each of NOR gates 1211-1214 receives either row address signal RA1 or a complementary row address signal /RA1, and either row address signal RA2 or a complementary row address signal /RA2. Each of NOR gates 1215-1218 receives either a row address signal RA3 or a complementary row address signal RA3, and either a row address signal RA4 or a complementary row address signal /RA4. Each of NOR gates 1211-1218 has its output signal provided to a word driver as a predecode signal via two inverters. For example, NOR gate 1211 receives row address signals /RA1 and /RA2 to provide predecode signal X1 to the word driver via two inverters 1221a-1231. Therefore, one of predecode signals X1-X4 attains an H level according to four combinations of row address signals RA1, /RA1, RA2, and /RA2. For example, predecode signal X1 attains an L level when row address signals /RA1 and /RA2 both attain an L level. Furthermore, according to the four combinations of row address signals RA3, /RA3, RA4 and /RA4, one of predecode signals X5-X8 attains an H level. For example, when row address signals /RA3 and /RA4 both attain an L level, predecode signal X5 is pulled up to an H level.
In order to carry out a stress test on word line WL and access transistor 112 in the above-described DRAM, boosted potential VPP must be supplied to word line WL only during a predetermined time period. However, the testing is time-consuming according to increase in the capacity of a memory, resulting in increase in the cost required for testing. For example, a reliability test called "burn-in" that applies acceleration stress on a memory cell had a problem that the testing time period becomes longer as the number of memory cells becomes greater. The stress testing on a gate oxide film in access transistor 112 and on a dielectric film in cell capacitor 113 are extremely important. However, the number of word lines n that can be activated at one time is limited in a normal operation. Therefore, the testing must be carried for N (total number of word lines) /n times in order to evaluate all the word lines. The time required for testing becomes longer in accordance with a higher integration density of a chip.
A method of increasing the number of word lines that are activated simultaneously is considered as one way of reducing the testing time. An example of an DRAM that allows testing by such a method is shown in FIG. 41. FIG. 41 is substantially identical to FIG. 3 in pp. 639-642 of "IEDM93, DIGEST". Referring to FIG. 41, the DRAM includes a memory cell array 11 having a plurality of word lines WL and bit lines (not shown) crossing thereto, a row decoder 12 selecting one of word lines WL, a column decoder 13 selecting one bit line, and a plurality of N channel MOS transistors 1 connected corresponding to word lines WL. Row decoder 12 includes a plurality of word drivers WD, each driving a corresponding word line WL. All transistors 1 are turned on in response to one multi-selection signal MLT, whereby boosted test potential VST is supplied to all word lines WL. Therefore, stress can be applied to all the access transistors simultaneously since all the word lines are activated in a burn-in mode. Thus, the testing time can be shortened.
FIG. 42 shows a semiconductor (silicon) wafer 70 not yet subjected to a dicing process. A plurality of semiconductor chips 71 are formed on silicon wafer 70. A probe card 2 as shown in FIG. 43 is used in carrying out an acceleration test such as burn-in on the wafer shown in FIG. 42. Probe card 2 includes an opening 3 corresponding to a plurality of chips 71 aligned in one row in the vertical direction (three in FIG. 42) in wafer 70. A plurality of probes 4 are provided corresponding to the three chips 71 at the periphery of opening 3. This probe card 2 is set on wafer 70 to carry out a burn-in test in a wafer state. Probes 4 are brought into contact with the pads (not shown) of the three chips 71, so that power supply and a signal for testing can be applied simultaneously to the three chips 71. In order to test all the chips 71 in wafer 70, the setting position of probe card 2 must be shifted 5 times.
Since boosted test potential VST is supplied to word line WL via transistor 1 in the DRAM of FIG. 41, boosted test potential VST must be higher than boosted potential VPP by the threshold voltage of transistor 1 in order to supply a boosted potential VPP higher than the power supply potential to word line WL. Therefore, a great stress is applied on transistor 1 controlled by multi-selection signal MLT which is greater than that on the access transistor that should actually receive the greater stress. As a result, there is a possibility that these transistors 1 become defective. Furthermore, since one transistor 1 must be provided corresponding to each word line WL, reduction in the pitch between word lines WL due to miniaturization of the chip is attributed to the difficulty of forming a transistor with such small pitches. As a result, there is a possibility that these transistors 1 become defective.
When silicon wafer 70 as shown in FIG. 42 is to be tested using probe card 2 shown in FIG. 43, the setting position of probe card 2 must be changed 5 times with respect to wafer 70. There was a problem that the entire testing of wafer 70 was time consuming.